aparapiベンチ

AparapiBench.java

package kishida.aparapisample;

import com.amd.aparapi.Kernel;
import java.util.Random;
import java.util.stream.IntStream;

/**
 *
 * @author kishida
 */
public class AparapiBench extends Kernel{
	static float[] in = new float[65536];
	static float[] filter = new float[11 * 11];
	static float[] out = new float[in.length];
	public static void main(String[] args) {
		Random r = new Random(1234);
		for(int i = 0; i < in.length; ++i){
			in[i] = r.nextFloat();
		}
		for(int i = 0; i < filter.length; ++i){
			filter[i] = r.nextFloat();
		}
		k = new AparapiSample(); // new Kernel(){}やnew AparapiSample(){}だとだめ
		k.setExplicit(true);
		bench("gpu", () -> gpu());
		bench("para", () -> para());
		bench("seq", () -> seq());
		System.out.println(k.getExecutionMode());
	}
	public static void gpu(){
		k.put(in);
		k.put(filter);
		k.execute(in.length - filter.length + 1);
		k.get(out);
	}

	public static void seq(){
		for(int id = 0; id < in.length - filter.length + 1; ++id){
			float result = 0;
			for(int i = 0; i < filter.length; ++i){
				result += filter[i] * in[id + i];
			}
			out[id] = result;;

		}
	}
	public static void para(){
		IntStream.range(0, in.length - filter.length).parallel().forEach(id -> {
			float result = 0;
			for(int i = 0; i < filter.length; ++i){
				result += filter[i] * in[id + i];
			}
			out[id] = result;;

		});
	}

	static Kernel k;
	@Override
	public void run() {
		int id = getGlobalId();

		float result = 0;
		for(int i = 0; i < 11 * 11; ++i){ // filter.lengthが使えない
			result += filter[i] * in[id + i];
		}
		out[id] = result;//result;
	}


	static void bench(String name, Runnable proc){
		for(int i = 0; i < 10; ++i){
			proc.run();
		}
		long start = System.currentTimeMillis();
		for(int i = 0; i < 1000; ++i){
			proc.run();
		}
		System.out.printf("%s:%.3fs%n", name, (System.currentTimeMillis() - start) / 1000.);
	}

}

mac

Mac book pro
i7 4コア 2.8GHz
Intel Iris pro 1024MB

gpu:8.025s
para:1.287s
seq:6.648s
GPU

mac JTP

gpu:2.836s
para:1.508s
seq:5.727s
JTP

win7 GT730

i7 2600K 3.4GHz
NVIDIA GT730

gpu:0.790s
para:1.293s
seq:7.326s
GPU

gpu:1.555s
para:1.185s
seq:7.354s
JTP

win7 GTX970

NVIDIA GTX970

gpu:0.486s
para:1.148s
seq:7.340s
GPU